JP4942860B2 - Recognition dictionary creation device, speech recognition device, and speech synthesis device - Google Patents

Description

  The present invention relates to a recognition dictionary creation device for registering vocabulary in a speech recognition dictionary using speech uttered by a user, a speech recognition device and a speech synthesis device using the recognition dictionary creation device.

  Depending on the application to which speech recognition is applied, the speech uttered by the user may be registered and used as a recognition target word. Hereinafter, this operation is referred to as user dictionary generation. As an example of generating a user dictionary by voice, a radio station name corresponding to a radio frequency may be registered by voice, or a person name or a place name corresponding to a telephone number may be registered by voice.

In addition, a language switching function is required for speech recognition installed in devices that can be used across multiple countries, such as car navigation systems and portable terminals.
As a related art, for example, in Patent Document 1, when switching the language used in an electronic dictionary, character data obtained by voice recognition of a voice spoken by a user is collated with a word stored in the apparatus. Thus, there is disclosed a method of switching the language used to determine the language that the user wants to use.

In general, voice data is collected for each language, and a voice uttered by a user is recognized using a voice recognition algorithm or a voice standard model constructed using the collected voice data. For this reason, when the language is switched, it is necessary to switch the voice recognition means itself or the voice standard model.
Conventionally, by using generally known speech recognition technology, for each language, a phoneme label string that best represents the speech generated by the user is generated and stored as a user dictionary. There has also been proposed a voice recognition device that can recognize a voice generated by a user even when a language used in the system is switched.

However, when a phoneme label string is created each time the language is changed, the speech is stored in the memory for processing, and thus there is a problem that a large-capacity memory that can secure a storage area for the speech is required. It was.
In addition, if the generated speech cannot be stored in the memory, it is necessary to create a phoneme label sequence for all possible languages, but even when creating a phoneme label sequence for a single language. Since a great deal of time is required, the processing time required for all possible languages is enormous. In addition to this, a large-capacity memory capable of storing phoneme label strings for all languages is also required.

  The present invention has been made to solve the above-described problems, and does not require a large-capacity memory for storing speech speech, and it is not necessary to create a phoneme label string in advance for all languages. An object of the present invention is to obtain a recognition dictionary creation device capable of shortening the creation time of each phoneme label string, and a speech recognition device and speech synthesis device using the recognition dictionary creation device.

JP 2001-282788 A

  The recognition dictionary creation apparatus according to the present invention includes an acoustic analysis unit that acoustically analyzes a speech signal of an input speech and outputs a time series of acoustic features, and an acoustic standard that stores acoustic standard patterns indicating standard acoustic features for each language. Acoustic data matching that creates a phoneme label string of input speech by collating the time series of the acoustic features of the input speech input from the pattern storage unit and the acoustic analysis unit with the acoustic standard pattern stored in the acoustic standard pattern storage unit A user dictionary storing a user dictionary in which the phoneme label sequence of the input speech created by the acoustic data matching unit is registered, a language storage unit storing the language of the phoneme label sequence registered in the user dictionary, A language switching unit for switching languages, a mapping table storage unit for storing a mapping table in which correspondence between phoneme labels between languages is defined, and mapping The phoneme label sequence registered in the user dictionary is converted from the phoneme level sequence stored in the language storage unit to the phoneme label sequence of the language switched by the language switching unit with reference to the mapping table stored in the table storage unit. A phoneme label string conversion unit.

According to the present invention, a user dictionary in which a phoneme label sequence of input speech is registered, and a mapping table in which a correspondence relationship between phoneme labels between languages is defined, and the phonemes registered in the user dictionary with reference to the mapping table are provided. The label string is converted from the phoneme label string in the language when the user dictionary is created into the phoneme label string in the language after switching.
Even if the language is switched in this way, the registered vocabulary can be quickly converted to the language after switching by referring to the mapping table, so that a large-capacity memory for storing the speech is not necessary, and all There is no need to create a phoneme label string in advance for each language, and the time for creating a phoneme label string for each language can be shortened.

It is a block diagram which shows the structure of the recognition dictionary creation apparatus by Embodiment 1 of this invention. 4 is a flowchart showing a flow of a user dictionary registration operation by the recognition dictionary creation device of the first embodiment. 6 is a flowchart showing a flow of user dictionary registration operation after language switching by the recognition dictionary creating apparatus of the first embodiment. It is a block diagram which shows the structure of the speech recognition apparatus by Embodiment 2 of this invention. 10 is a flowchart showing a flow of operations performed by the speech recognition apparatus according to the second embodiment. It is a block diagram which shows the structure of the speech synthesizer by Embodiment 3 of this invention. 10 is a flowchart illustrating a flow of operations performed by the speech synthesizer according to the third embodiment. It is a block diagram which shows the structure of the recognition dictionary creation apparatus by Embodiment 4 of this invention. 10 is a flowchart illustrating a flow of a user dictionary registration operation performed by the recognition dictionary creation device according to the fourth embodiment. 15 is a flowchart showing a flow of user dictionary registration operation after language switching by the recognition dictionary creating apparatus of the fourth embodiment.

Hereinafter, in order to explain the present invention in more detail, modes for carrying out the present invention will be described with reference to the accompanying drawings.
Embodiment 1 FIG.
FIG. 1 is a block diagram showing a configuration of a recognition dictionary creation apparatus according to Embodiment 1 of the present invention. In FIG. 1, a recognition dictionary creating apparatus 1 according to Embodiment 1 includes a microphone 2a, a voice capturing unit 2, an acoustic analysis unit 3, an acoustic standard pattern 4 for each language, an acoustic data matching unit 5, a user dictionary registration unit (user dictionary (Storage unit) 6, user dictionary creation language storage unit (language storage unit) 7, language switching unit 8, phoneme label string conversion unit 9, and inter-language acoustic data mapping table storage unit (mapping table storage unit) 10.

  The sound capturing unit 2 is a component that converts sound captured by the microphone 2a into a digital signal. The acoustic analysis unit 3 is a configuration unit that analyzes the voice signal converted into a digital signal by the voice capturing unit 2 and converts it into a time series of acoustic features. For example, an audio signal is analyzed at regular time intervals, and an acoustic feature amount (acoustic feature amount vector) representing the feature of the speech is calculated.

  The acoustic standard pattern 4 is a standard acoustic feature corresponding to each phoneme label string of the language X (X = 1, 2, 3,...) (A standard model representing the characteristics of acoustic features for speech fragments). For example, it is modeled by HMM (Hidden Markov Model) etc. with phonemes as units. The acoustic data matching unit 5 matches the time series of the acoustic features of the input speech obtained by the acoustic analysis unit 3 with the acoustic standard pattern 4 of the language X and corresponds to the standard acoustic features constituting the acoustic standard pattern 4. This is a component that creates a phoneme label string most similar to the input speech from the phoneme label string that has been processed.

  The user dictionary registration unit 6 is a component having a user dictionary, and stores the phoneme label string of the input speech created by the acoustic data matching unit 5 in the user dictionary. The user dictionary creation language storage unit 7 is a storage unit that stores a set language set as the speech recognition language when the user dictionary is created. The language switching unit 8 is a configuration unit that switches a setting language used as a language for speech recognition.

  The phoneme label string conversion unit 9 uses the inter-language acoustic data mapping table to convert the phoneme label string expressed in the language registered in the user dictionary to the phoneme label string of the language changed by the language switching unit 8. It is the component part to convert. The inter-language acoustic data mapping table storage unit 10 is a storage unit that stores an inter-language acoustic data mapping table indicating a correspondence relationship between pairs of different languages and phoneme labels of these languages.

  If one language cannot express the phoneme label of the other language, a similar phoneme label is associated among the phoneme labels that can be expressed in the other language. For example, Japanese phoneme label / l / cannot be expressed in Japanese. Therefore, the Japanese phoneme label / r / whose pronunciation is similar to the English phoneme label / l / is associated with the inter-language acoustic data mapping table in Japanese and English.

  Also, a voice capturing unit 2, an acoustic analysis unit 3, an acoustic standard pattern 4, an acoustic data matching unit 5, a user dictionary registration unit 6, a user dictionary creation language storage unit 7, a language switching unit 8, a phoneme label string conversion unit 9, and The inter-language acoustic data mapping table storage unit 10 stores a recognition dictionary creation program according to the gist of the present invention in a computer and causes the CPU to execute the computer as a specific means in which hardware and software cooperate. Can be realized above. Further, the storage area used in the acoustic standard pattern 4, the user dictionary registration unit 6, the user dictionary creation language storage unit 7 and the inter-language acoustic data mapping table storage unit 10 is a storage device mounted on the computer, such as a hard disk device or the like. Built on external storage media.

Next, the operation will be described.
FIG. 2 is a flowchart showing the flow of the user dictionary registration operation by the recognition dictionary creating apparatus of the first embodiment.
After the user gives an instruction to start creating a user dictionary using the input device (step ST1), the user speaks the vocabulary to be registered. For example, it is assumed that the personal name “Michael” is spoken. The voice capturing unit 2 captures the voice uttered by the user via the microphone 2a, converts the input voice into a digital signal, and outputs the digital signal to the acoustic analysis unit 3 (step ST2).

  Subsequently, the user dictionary creation language storage unit 7 checks the language currently set in the acoustic data matching unit 5 at the time of user dictionary registration (step ST3) and registers it in itself (step ST4). The set language is a language that is set in advance as a language for speech recognition or speech synthesis in the speech recognition device or speech synthesis device using the recognition dictionary creation device 1. In the example of FIG. 2, English is set as the setting language. The acoustic analysis unit 3 acoustically analyzes the voice signal input from the voice capturing unit 2 in step ST2, and converts the voice signal into a time series of acoustic features (step ST5).

  The acoustic data matching unit 5 reads out the acoustic standard pattern 4 corresponding to the language (setting language) set in itself, and the acoustic standard pattern 4 of the setting language and the acoustic of the input speech obtained by the acoustic analysis unit 3 The optimal phoneme label representing the input speech most similar to the time series of the acoustic features of the input speech from the phoneme label sequence corresponding to the standard acoustic features constituting the acoustic standard pattern 4 by collating with the feature time series A column is created (step ST6). For example, when the input voice is “Michael” and the set language is English, “#, / m /, / a /, / i /, / k /, / l /, #” as shown in FIG. Is obtained.

  The user dictionary registration unit 6 registers the phoneme label string of the input speech created by the acoustic data matching unit 5 in the user dictionary (step ST7). As a result, a user dictionary in which a phoneme label string corresponding to the registered vocabulary text in the set language is registered is created.

Next, the operation when the set language is switched will be described.
FIG. 3 is a flowchart showing the flow of user dictionary registration operation after language switching by the recognition dictionary creation apparatus of the first embodiment, and shows a case where the language is switched after the user dictionary registration shown in FIG. 2 is executed. Show.
For example, when the user designates a new language in the language switching unit 8 using the input device, the language switching unit 8 sets the switched language in the phoneme label string conversion unit 9 (step ST1a). Here, it is assumed that the language has been switched to Japanese.
The phoneme label string conversion unit 9 reads the language stored in the user dictionary creation language storage unit 7 and checks the set language when registering the user dictionary (step ST2a). As described above, in FIG. 2, the setting language at the time of user dictionary registration is English.

  Subsequently, the phoneme label string conversion unit 9 uses the language set at the time of registration of the user dictionary confirmed in step ST2a and the language after switching designated by the language switching unit 8 to store the inter-language acoustic data mapping table. 10 is read, and the inter-language acoustic data mapping table corresponding to the language set at the time of registration of the user dictionary and the language after switching is read.

  As shown in FIG. 3, the inter-language acoustic data mapping table is table data indicating the correspondence between English phoneme labels and Japanese phoneme labels. For example, in FIG. 3, among the phoneme labels in English, three different phoneme labels with similar pronunciations indicated by the symbol A include those that cannot be expressed in Japanese. In this case, of the Japanese phoneme labels, one phoneme label (/ a /) similar to the pronunciation of the phoneme label indicated by symbol A is associated. Also, since the Japanese phoneme label / l / cannot be expressed in Japanese, the Japanese phoneme label / r / whose pronunciation is similar to the English phoneme label / l / is associated.

The phoneme label string conversion unit 9 converts the phoneme label string registered in the user dictionary based on the interlanguage acoustic data mapping table read from the interlanguage acoustic data mapping table storage unit 10 to the phoneme label string of the language after switching. (Step ST3a).
For example, as shown in FIG. 3, “#, / m /, / a /, / i /, / k /, / l /, #”, which is an English phoneme label string “Michael”, is English and Japanese. Based on the correspondence relationship in the inter-language acoustic data mapping table of words, it is converted into “#, / m /, / a /, / i /, / k /, / r /, #” which is a Japanese phoneme label string. Is done.
The method for creating the inter-language acoustic data mapping table is disclosed, for example, in Reference Document 1 below.
(Reference Document 1); Japanese Unexamined Patent Application Publication No. 2007-155833

  The user dictionary registration unit 6 re-stores the phoneme label sequence converted by the phoneme label sequence conversion unit 9 in step ST3a in the user dictionary (step ST4a). In FIG. 3, since the registered vocabulary is “Michael” and the language after switching is Japanese, “#, / m /, / a /, / i /, / k” is a Japanese phoneme label string. /, / R /, # "are stored as one registered word.

As described above, according to the first embodiment, the user dictionary in which the phoneme label sequence of the input speech is registered, and the inter-language acoustic data mapping table in which the correspondence of phoneme labels between languages is defined, With reference to the inter-acoustic data mapping table, the phoneme label string registered in the user dictionary is converted from the phoneme label string in the language when the user dictionary is created into the phoneme label string in the language after switching.
By configuring in this way, even if the set language is changed from the time of registration of the user dictionary, it is possible to convert the phoneme label string based on the inter-language acoustic data mapping table and to change the language user after the change. A dictionary can be created, and the processing time for creating a phoneme label string of a corresponding language can be significantly reduced.
Also, even if a phoneme label string is created every time the language is changed, it is not necessary to save the utterance speech, only the phoneme label string at the time of registering the user dictionary is saved, and the phonemes for all assumed languages are saved. There is no need to prepare labels in advance. This eliminates the need for a large-capacity memory.

Embodiment 2. FIG.
FIG. 4 is a block diagram showing a configuration of a speech recognition apparatus according to Embodiment 2 of the present invention, and shows a speech recognition apparatus using the recognition dictionary creation apparatus according to Embodiment 1 described above. 4, in addition to the configuration of the recognition dictionary creation device 1 shown in the first embodiment, the speech recognition device 1A according to the second embodiment includes a dictionary collation unit 11 and a general dictionary expressed by an acoustic standard pattern for each language. 12 and a recognition result output unit 13. In FIG. 4, the same reference numerals are given to components operating in the same or similar manner as in FIG. 1, and description thereof is omitted.

  The dictionary collation unit 11 collates the phoneme label string of the input speech, the vocabulary of the general dictionary 12 expressed by the acoustic standard pattern of the set language, and the vocabulary registered in the user dictionary of the user dictionary registration unit 6, It is a component that identifies the vocabulary most similar to the phoneme label string of the input speech from the vocabulary of the general dictionary 12 and the user dictionary. The general dictionary 12 is a dictionary expressed by an acoustic standard pattern of a language X (X = 1, 2, 3,...), And a large vocabulary (phoneme label string) such as a place name of the language is registered. The recognition result output unit 13 is a component that outputs a speech recognition result, and outputs a vocabulary most similar to the phoneme label string of the input speech obtained as a result of collation by the dictionary collation unit 11.

  In addition, the dictionary collation unit 11, the general dictionary 12 expressed by the acoustic standard pattern for each language, and the recognition result output unit 13 store a speech recognition program according to the gist of the present invention in a computer and cause the CPU to execute the hardware recognition program. As a specific means in which the software and the software cooperate, it can be realized on the computer. Furthermore, the storage area used for the acoustic standard pattern 4 and the general dictionary 12 is constructed in a storage device mounted on the computer, such as a hard disk device or an external storage medium.

Next, the operation will be described.
FIG. 5 is a flowchart showing a flow of operations performed by the speech recognition apparatus according to the second embodiment.
After the user instructs the start of voice recognition using the input device (step ST1b), the user utters the voice that is the target of voice recognition. For example, it is assumed that the personal name “Michael” is spoken. The voice capturing unit 2 captures voice uttered by the user via the microphone 2a, converts the input voice into a digital signal, and outputs the digital signal to the acoustic analysis unit 3 (step ST2b). The acoustic analysis unit 3 acoustically analyzes the voice signal input from the voice capturing unit 2 in step ST2b, and converts the voice signal into a time series of acoustic features.

The acoustic data matching unit 5 reads the language stored in the user dictionary creation language storage unit 7 and confirms the set language when registering the user dictionary (step ST3b). In FIG. 5, it is assumed that the setting language at the time of user dictionary registration is Japanese.
Subsequently, the acoustic data matching unit 5 creates a phoneme label string of the set language for the input speech from the time series of the acoustic features of the input speech captured from the acoustic analysis unit 3 and the acoustic standard pattern 4 of the set language ( Step ST4b). For example, when the input speech is “Michael” and the setting language is Japanese, “#, / m /, / a /, / i /, / k /, / r /, # ".

  Next, the dictionary collation unit 11 includes a phoneme label string of the input speech created by the acoustic data matching unit 5, a vocabulary of the general dictionary 12 expressed by the acoustic standard pattern 4 of the set language, and a user dictionary registration unit 6. The vocabulary registered in the user dictionary is collated, and the vocabulary most similar to the phoneme label string of the input speech is specified from the vocabulary of the general dictionary 12 and the user dictionary (step ST5b). The recognition result output unit 13 outputs the vocabulary most similar to the phoneme label string of the input speech obtained as a result of collation by the dictionary collation unit 11 (step ST6b).

  As shown in FIG. 5, a large vocabulary such as a place name is registered as a phoneme label string in the general dictionary 12 expressed by an acoustic standard pattern of a set language (here, Japanese). In the user dictionary, as shown in the first embodiment, an arbitrary vocabulary is registered as a phoneme label string by the user's utterance. Here, when “#, / m /, / a /, / i /, / k /, / r /, #” is registered as the registered word 1 of the user dictionary, the dictionary collation unit 11 performs the input speech. The registered word 1 is specified as the vocabulary most similar to the phoneme label string of, and the recognition result output unit 13 outputs the registered word 1 as the recognition result.

  As described above, according to the second embodiment, in addition to the configuration of the recognition dictionary creating apparatus of the first embodiment, the general dictionary storage unit that stores the general dictionary 12 and the acoustic data matching unit 5 are used. A dictionary collation unit 11 that collates the phoneme label sequence of the input speech, the general dictionary 12 and the user dictionary, and identifies the vocabulary most similar to the phoneme label sequence of the input speech from the general dictionary 12 and the user dictionary; And the recognition result output unit 13 for outputting the vocabulary specified by the dictionary collation unit 11 as a speech recognition result. In addition to the effects of the first embodiment, the speech for performing speech recognition using the user dictionary The recognition device 1A can be provided.

Embodiment 3 FIG.
FIG. 6 is a block diagram showing a configuration of a speech synthesizer according to Embodiment 3 of the present invention, and shows a speech synthesizer using the recognition dictionary creating apparatus according to Embodiment 1 described above. In FIG. 6, the speech synthesizer 1B according to the third embodiment is expressed by the configuration of the recognition dictionary creation device 1 shown in the first embodiment and the acoustic standard pattern for each language shown in the second embodiment. In addition to the general dictionary 12, a text input unit 14, a registered word part detection unit 15, a registered word partial phoneme label string replacement unit (registered vocabulary replacement unit) 16, and a phoneme label string replacement unit (general dictionary replacement unit) 17 of other parts. And a speech synthesizer 18. In FIG. 6, the same reference numerals are given to components that operate in the same or similar manner as in FIGS. 1 and 4, and description thereof is omitted.

  The text input unit 14 is a component that inputs text to be converted into speech. The registered word part detection unit 15 is a component that detects a registered word registered in the user dictionary from the input text captured from the text input unit 14. The registered word partial phoneme label string replacement unit 16 is a component that replaces the registered word detected by the registered word part detection unit 15 with the phoneme label string imported from the user dictionary. The other part phoneme label string replacement unit 17 is a component that inputs an input text part other than the registered word detected by the registered word part detection unit 15 via the registered word part phoneme label string replacement unit 16. The words in the input text portion other than the registered words are replaced with phoneme label strings taken from the general dictionary 12 expressed in the acoustic standard pattern of the set language. The speech synthesizer 18 is a component that generates synthesized speech of the input text from the phoneme label string for the input text obtained by the phoneme label string replacement units 16 and 17.

  Note that the text input unit 14, the registered word part detection unit 15, the registered word part phoneme label string replacement unit 16, the other part phoneme label string replacement unit 17 and the speech synthesis unit 18 execute a speech synthesis program according to the spirit of the present invention. By being stored in a computer and executed by a CPU, it can be realized on the computer as a specific means in which hardware and software cooperate. Furthermore, the storage area used for the acoustic standard pattern 4 and the general dictionary 12 is constructed in a storage device mounted on the computer, such as a hard disk device or an external storage medium.

Next, the operation will be described.
FIG. 7 is a flowchart showing a flow of operations performed by the speech synthesizer according to the third embodiment.
The user uses the text input unit 14 to input text to be converted into speech (step ST1c). At this time, an identifier for identifying a registered word in the user dictionary is set. For example, as shown in FIG. 7, when the registered word 1 of the user dictionary is input as text, double parentheses that are registered word identifiers are set before and after the registered word 1.

The registered word part detection unit 15 takes in the input text from the text input unit 14, and detects a registered word using the registered word identifier set in the input text (step ST2c). In the example of FIG. 7, registered word 1 in which double parentheses are set before and after is detected.
Next, the registered word partial phoneme label string replacement unit 16 replaces the registered word detected by the registered word part detection unit 15 with the phoneme label string imported from the user dictionary (step ST3c). As a result, the registered word 1 is replaced with “#, / m /, / a /, / i /, / k /, / r /, #” which is the corresponding phoneme label string.

  The phoneme label string replacement unit 17 of the other part inputs a part other than the registered word detected by the registered word part detection unit 15 in the input text via the registered word partial phoneme label string replacement unit 16. The words in the input text part other than are replaced with the phoneme label string imported from the general dictionary 12 of the set language (step ST4c). Here, it is assumed that the setting language is Japanese, and the input text part other than the registered word is the particle “ha”, the noun “Osaka”, the particle “ni”, and the verb “ni”. As shown in FIG. 7, the corresponding phoneme label string registered in the Japanese general dictionary 12 is replaced.

  The speech synthesizer 18 generates synthesized speech of the input text from the phoneme label sequence for the input text obtained by the registered word partial phoneme label sequence replacement unit 16 and the phoneme label sequence replacement unit 17 of the other part (step) ST5c). In the example of FIG. 7, the synthesized voice “Micle went to Osaka” is output. Here, the parts other than the registered word 1 are uttered by Japanese phoneme labels, but the registered word 1 “Mikel” is set in the user dictionary as shown in the first embodiment. Because it is English, it becomes an English utterance.

As described above, according to the third embodiment, in addition to the configuration of the recognition dictionary creating apparatus of the first embodiment, the text input unit 14 for inputting text and the text input from the text input unit 14 A registered word part detection unit 15 that detects a vocabulary part corresponding to a phoneme label string registered in the user dictionary from the character string, and a corresponding phoneme obtained from the user dictionary by the vocabulary part detected by the registered word part detection unit 15 A part other than the vocabulary part detected by the registered word part detection unit 15 in the registered word part phoneme label string replacement part 16 to be replaced with the label string and a character string of the text is replaced with a corresponding phoneme label string in the general dictionary 12. The other part phoneme label string replacement unit 17, the registered word partial phoneme label string replacement unit 16 and the other part phoneme label string replacement unit 17. Therefore the phoneme label string of the text obtained, provided with a voice synthesizer 18 for generating synthesized speech of the text.
With this configuration, in addition to the effects of the first embodiment, it is possible to provide a speech synthesizer 1B that performs speech synthesis using a user dictionary.

Embodiment 4 FIG.
FIG. 8 is a block diagram showing a configuration of a recognition dictionary creation apparatus according to Embodiment 4 of the present invention. In FIG. 8, the recognition dictionary creation apparatus 1a of the fourth embodiment includes a registration acoustic pattern setting unit 19 instead of the user dictionary creation language storage unit 7 in the configuration of the first embodiment. The registration acoustic pattern setting unit 19 is the language of the acoustic standard pattern 4 used for the processing of the acoustic data matching unit 5 regardless of the setting language set in the speech recognition device or speech synthesis device using the recognition dictionary creation device 1a. As a configuration unit for setting a predetermined language registered in advance in itself. This predetermined language is registered in advance in the acoustic pattern setting unit 19 at the time of registration regardless of the set language. In FIG. 8, the same or similar components as those shown in FIG.

Next, the operation will be described.
FIG. 9 is a flowchart showing the flow of a user dictionary registration operation by the recognition dictionary creation apparatus of the fourth embodiment.
After the user instructs the user dictionary creation start using the input device (step ST1d), the user speaks the vocabulary to be registered. For example, it is assumed that the personal name “Michael” is spoken. The voice capturing unit 2 captures the voice uttered by the user via the microphone 2a, converts the input voice into a digital signal, and outputs the digital signal to the acoustic analysis unit 3 (step ST2d).

  Subsequently, the registered acoustic pattern setting unit 19 sets a predetermined language registered in advance in the acoustic data matching unit 5 instead of the system setting language (step ST3d). In the example of FIG. 9, English is the predetermined language. The acoustic analysis unit 3 acoustically analyzes the voice signal input from the voice capturing unit 2 in step ST2d, and converts the voice signal into a time series of acoustic features (step ST4d).

  The acoustic data matching unit 5 reads the acoustic standard pattern 4 corresponding to the predetermined language set from the registration acoustic pattern setting unit 19, and the acoustic standard pattern 4 of the set language and the input speech obtained by the acoustic analysis unit 3. An optimal phoneme label string representing the input speech is created from the time series of the acoustic features (step ST5d). When the input voice is “Michael” and the predetermined language is English, as shown in FIG. 9, it is called “#, / m /, / a /, / i /, / k /, / l /, #”. A phoneme label string is obtained.

The user dictionary registration unit 6 registers the phoneme label string of the input speech created by the acoustic data matching unit 5 in the user dictionary (step ST6d).
Next, based on the interlanguage acoustic data mapping table read by the phoneme label string conversion unit 9 from the interlanguage acoustic data mapping table storage unit 10, a predetermined language for the input speech (registered vocabulary) obtained as described above is used. The phoneme label string of the set language and the phoneme label of the set language currently set in the system, and the phoneme label string of the registered vocabulary in the predetermined language registered in the user dictionary is converted into the phoneme label string of the set language, The current user dictionary is registered in the user dictionary registration unit 6 (step ST7d).

Next, the operation when the set language is switched will be described.
FIG. 10 is a flowchart showing the flow of user dictionary registration operation after language switching by the recognition dictionary creating apparatus of the fourth embodiment. The case where the language is switched after the user dictionary registration shown in FIG. 9 is executed. Show.
When the user designates a new language in the language switching unit 8 using the input device, the language switching unit 8 sets the language after switching in the phoneme label string conversion unit 9 (step ST1e). Here, it is assumed that the language has been switched to Japanese.

The phoneme label string conversion unit 9 searches the inter-language acoustic data mapping table storage unit 10 using the switched language designated by the language switching unit 8 and the predetermined language, and performs predetermined processing when registering the user dictionary. The inter-language acoustic data mapping table corresponding to the language and the language after switching is read, and based on this inter-language acoustic data mapping table, the phoneme label string of the predetermined language registered in the user dictionary is converted into the phoneme label string of the language after switching. (Step ST2e).
For example, the phoneme label string “#, / m /, / a /, / i /, / k /, / l /, #” of English “Michael” that is a predetermined language is Japanese after switching. Is converted into “#, / m /, / a /, / i /, / k /, / r /, #”, which is a Japanese phoneme label string Is done.

  The user dictionary registration unit 6 additionally stores the phoneme label string converted by the phoneme label string conversion unit 9 in step ST2e in the user dictionary (step ST3e). In FIG. 10, since the registered vocabulary text is “Michael” and the language after switching is Japanese, “#, / m /, / a /, / i /, / k /, / r /, # "are stored as registered words.

As described above, according to the fourth embodiment, the user dictionary in which the phoneme label sequence of the input speech is registered, the inter-language acoustic data mapping table in which the correspondence between phoneme labels between languages is defined, and the acoustic standard pattern A registration-time acoustic pattern setting unit 19 for selecting a preset acoustic standard pattern of a language, and referring to an inter-language acoustic data mapping table to register a phoneme label string registered in a user dictionary The phoneme label string of the language selected by the acoustic pattern setting unit 19 is converted into the phoneme label string of the language after switching.
With this configuration, in the first embodiment, when N languages can be set as target languages of the vocabulary to be registered in the user dictionary, the languages at the time of registration in the user dictionary and the languages that can be set are set. When the inter-language acoustic data mapping table corresponding to all combinations of (N × (N−1)) / 2 is required, one predetermined language set by the acoustic pattern setting unit 19 at the time of registration And an inter-language acoustic data mapping table corresponding to (N-1) combinations of the settable languages, and the data size of the inter-language acoustic data mapping table can be reduced.

  In the second embodiment and the third embodiment, the case where the speech recognition device and the speech synthesis device are configured using the recognition dictionary creation device 1 according to the first embodiment has been described. In the configuration shown in FIG. 8, instead of the recognition dictionary creation device according to the first embodiment, a speech recognition device and a speech synthesis device are configured by combining the recognition dictionary creation device 1a according to the fourth embodiment shown in FIG. It doesn't matter. As a result, it is possible to provide a speech recognition apparatus and a speech synthesis apparatus that can also obtain the effects of the fourth embodiment.

  The recognition dictionary creation apparatus according to the present invention does not require a large-capacity memory for storing uttered speech, eliminates the need to create phoneme label sequences in advance for all languages, and shortens the creation time of phoneme label sequences for each language. Therefore, it is suitable for a speech recognition device and a speech synthesis device for in-vehicle devices.

Claims (6)

An acoustic analysis unit that acoustically analyzes a voice signal of an input voice and outputs a time series of acoustic features;
An acoustic standard pattern storage unit for storing acoustic standard patterns indicating standard acoustic features for each language;
An acoustic data matching unit that creates a phoneme label string of the input speech by collating a time series of acoustic features of the input speech input from the acoustic analysis unit and an acoustic standard pattern stored in the acoustic standard pattern storage unit When,
A user dictionary storage unit that stores a user dictionary in which the phoneme label sequence of the input speech created by the acoustic data matching unit is registered;
A language storage unit for storing the language of the phoneme label sequence registered in the user dictionary;
A language switching unit for switching languages;
A mapping table storage unit that stores a mapping table in which correspondences between phoneme labels between languages are defined;
Referring to the mapping table stored in the mapping table storage unit, the phoneme label sequence registered in the user dictionary is changed from the phoneme level sequence of the language stored in the language storage unit to the language switched by the language switching unit. A recognition dictionary creation device comprising a phoneme label string conversion unit for converting into a phoneme label string. An acoustic analysis unit that acoustically analyzes a voice signal of an input voice and outputs a time series of acoustic features;
An acoustic standard pattern storage unit for storing acoustic standard patterns indicating standard acoustic features for each language;
An acoustic data matching unit that creates a phoneme label string of the input speech by collating a time series of acoustic features of the input speech input from the acoustic analysis unit and an acoustic standard pattern stored in the acoustic standard pattern storage unit When,
A user dictionary storage unit that stores a user dictionary in which the phoneme label sequence of the input speech created by the acoustic data matching unit is registered;
A language storage unit for storing the language of the phoneme label sequence registered in the user dictionary;
A language switching unit for switching languages;
A mapping table storage unit that stores a mapping table in which correspondences between phoneme labels between languages are defined;
With reference to the mapping table stored in the mapping table storage unit, the phoneme label sequence registered in the user dictionary is changed from the phoneme label sequence of the language stored in the language storage unit to the language switched by the language switching unit. A phoneme label string conversion unit for converting to a phoneme label string;
A general dictionary storage unit for storing a general dictionary of vocabulary expressed in the acoustic standard pattern;
The phoneme label sequence of the input speech is selected from the general dictionary and the user dictionary by collating the phoneme label sequence of the input speech created by the acoustic data matching unit, the general dictionary, and the user dictionary. A dictionary matching unit that identifies the vocabulary most similar to
A speech recognition apparatus comprising: a recognition result output unit that outputs the vocabulary specified by the dictionary collation unit as a speech recognition result. An acoustic analysis unit that acoustically analyzes a voice signal of an input voice and outputs a time series of acoustic features;
An acoustic standard pattern storage unit for storing acoustic standard patterns indicating standard acoustic features for each language;
An acoustic data matching unit that creates a phoneme label string of the input speech by collating a time series of acoustic features of the input speech input from the acoustic analysis unit and an acoustic standard pattern stored in the acoustic standard pattern storage unit When,
A user dictionary storage unit that stores a user dictionary in which the phoneme label sequence of the input speech created by the acoustic data matching unit is registered;
A language storage unit for storing the language of the phoneme label sequence registered in the user dictionary;
A language switching unit for switching languages;
A mapping table storage unit that stores a mapping table in which correspondences between phoneme labels between languages are defined;
With reference to the mapping table stored in the mapping table storage unit, the phoneme label sequence registered in the user dictionary is changed from the phoneme label sequence of the language stored in the language storage unit to the language switched by the language switching unit. A phoneme label string conversion unit for converting to a phoneme label string;
A text input section for entering text;
A registered word part detection unit for detecting a vocabulary part corresponding to a phoneme label string registered in the user dictionary from a character string of text input from the text input unit;
A registered vocabulary replacement unit that replaces the vocabulary part detected by the registered word part detection unit with a phoneme label string corresponding to the vocabulary part acquired from the user dictionary;
A general dictionary replacement unit that replaces a part of the text string other than the vocabulary part detected by the registered word part detection unit with a phoneme label string of a corresponding vocabulary of the general dictionary;
A speech synthesizer comprising: a speech synthesizer that generates synthesized speech of the text from the phoneme label string of the text obtained by the registered vocabulary replacement unit and the general dictionary replacement unit. An acoustic analysis unit that acoustically analyzes a voice signal of an input voice and outputs a time series of acoustic features;
An acoustic standard pattern storage unit for storing acoustic standard patterns indicating standard acoustic features for each language;
An acoustic standard pattern setting unit for selecting an acoustic standard pattern in a preset language from the acoustic standard patterns stored in the acoustic standard pattern storage unit;
Acoustic data matching for generating a phoneme label string of the input speech by collating a time series of acoustic features of the input speech input from the acoustic analysis unit and an acoustic standard pattern of a language selected by the acoustic standard pattern setting unit And
A user dictionary storage unit that stores a user dictionary in which the phoneme label sequence of the input speech created by the acoustic data matching unit is registered;
A language switching unit for switching languages;
A mapping table storage unit that stores a mapping table in which correspondences between phoneme labels between languages are defined;
With reference to the mapping table stored in the mapping table storage unit, the phoneme label sequence registered in the user dictionary is switched by the language switching unit from the phoneme label sequence of the language selected by the acoustic standard pattern setting unit. Recognition dictionary creating device comprising a phoneme label string conversion unit for converting into a phoneme label string of a different language. An acoustic analysis unit that acoustically analyzes a voice signal of an input voice and outputs a time series of acoustic features;
An acoustic standard pattern storage unit for storing acoustic standard patterns indicating standard acoustic features for each language;
An acoustic standard pattern setting unit for selecting an acoustic standard pattern in a preset language from the acoustic standard patterns stored in the acoustic standard pattern storage unit;
Acoustic data matching for generating a phoneme label string of the input speech by collating a time series of acoustic features of the input speech input from the acoustic analysis unit and an acoustic standard pattern of a language selected by the acoustic standard pattern setting unit And
A user dictionary storage unit that stores a user dictionary in which the phoneme label sequence of the input speech created by the acoustic data matching unit is registered;
A language switching unit for switching languages;
A mapping table storage unit that stores a mapping table in which correspondences between phoneme labels between languages are defined;
With reference to the mapping table stored in the mapping table storage unit, the phoneme label sequence registered in the user dictionary is switched by the language switching unit from the phoneme label sequence of the language selected by the acoustic standard pattern setting unit. A phoneme label string conversion unit for converting to a phoneme label string of a different language;
A general dictionary storage unit for storing a general dictionary of vocabulary expressed in the acoustic standard pattern;
The phoneme label sequence of the input speech is selected from the general dictionary and the user dictionary by collating the phoneme label sequence of the input speech created by the acoustic data matching unit, the general dictionary, and the user dictionary. A dictionary matching unit that identifies the vocabulary most similar to
A speech recognition apparatus comprising: a recognition result output unit that outputs the vocabulary specified by the dictionary collation unit as a speech recognition result. An acoustic analysis unit that acoustically analyzes a voice signal of an input voice and outputs a time series of acoustic features;
An acoustic standard pattern storage unit for storing acoustic standard patterns indicating standard acoustic features for each language;
An acoustic standard pattern setting unit for selecting an acoustic standard pattern in a preset language from the acoustic standard patterns stored in the acoustic standard pattern storage unit;
Acoustic data matching for generating a phoneme label string of the input speech by collating a time series of acoustic features of the input speech input from the acoustic analysis unit and an acoustic standard pattern of a language selected by the acoustic standard pattern setting unit And
A user dictionary storage unit that stores a user dictionary in which the phoneme label sequence of the input speech created by the acoustic data matching unit is registered;
A language switching unit for switching languages;
A mapping table storage unit that stores a mapping table in which correspondences between phoneme labels between languages are defined;
With reference to the mapping table stored in the mapping table storage unit, the phoneme label sequence registered in the user dictionary is switched by the language switching unit from the phoneme label sequence of the language selected by the acoustic standard pattern setting unit. A phoneme label string conversion unit for converting to a phoneme label string of a different language;
A text input section for entering text;
A registered word part detection unit for detecting a vocabulary part corresponding to a phoneme label string registered in the user dictionary from a character string of text input from the text input unit;
A registered vocabulary replacement unit that replaces the vocabulary part detected by the registered word part detection unit with a phoneme label string corresponding to the vocabulary part acquired from the user dictionary;
A general dictionary replacement unit that replaces a part of the text string other than the vocabulary part detected by the registered word part detection unit with a phoneme label string of a corresponding vocabulary of the general dictionary;
A speech synthesizer comprising: a speech synthesizer that generates synthesized speech of the text from the phoneme label string of the text obtained by the registered vocabulary replacement unit and the general dictionary replacement unit.

Images